s390/kernel/vdso.c

0001 // SPDX-License-Identifier: GPL-2.0
0002 /*
0003  * vdso setup for s390
0004  *
0005  *  Copyright IBM Corp. 2008
0006  *  Author(s): Martin Schwidefsky (schwidefsky@de.ibm.com)
0007  */
0008
0009 #include <linux/binfmts.h>
0010 #include <linux/compat.h>
0011 #include <linux/elf.h>
0012 #include <linux/errno.h>
0013 #include <linux/init.h>
0014 #include <linux/kernel.h>
0015 #include <linux/mm.h>
0016 #include <linux/slab.h>
0017 #include <linux/smp.h>
0018 #include <linux/time_namespace.h>
0019 #include <linux/random.h>
0020 #include <vdso/datapage.h>
0021 #include <asm/vdso.h>
0022
0023 extern char vdso64_start[], vdso64_end[];
0024 extern char vdso32_start[], vdso32_end[];
0025
0026 static struct vm_special_mapping vvar_mapping;
0027
0028 static union {
0029     struct vdso_data    data[CS_BASES];
0030     u8          page[PAGE_SIZE];
0031 } vdso_data_store __page_aligned_data;
0032
0033 struct vdso_data *vdso_data = vdso_data_store.data;
0034
0035 enum vvar_pages {
0036     VVAR_DATA_PAGE_OFFSET,
0037     VVAR_TIMENS_PAGE_OFFSET,
0038     VVAR_NR_PAGES,
0039 };
0040
0041 #ifdef CONFIG_TIME_NS
0042 struct vdso_data *arch_get_vdso_data(void *vvar_page)
0043 {
0044     return (struct vdso_data *)(vvar_page);
0045 }
0046
0047 static struct page *find_timens_vvar_page(struct vm_area_struct *vma)
0048 {
0049     if (likely(vma->vm_mm == current->mm))
0050         return current->nsproxy->time_ns->vvar_page;
0051     /*
0052      * VM_PFNMAP | VM_IO protect .fault() handler from being called
0053      * through interfaces like /proc/$pid/mem or
0054      * process_vm_{readv,writev}() as long as there's no .access()
0055      * in special_mapping_vmops().
0056      * For more details check_vma_flags() and __access_remote_vm()
0057      */
0058     WARN(1, "vvar_page accessed remotely");
0059     return NULL;
0060 }
0061
0062 /*
0063  * The VVAR page layout depends on whether a task belongs to the root or
0064  * non-root time namespace. Whenever a task changes its namespace, the VVAR
0065  * page tables are cleared and then they will be re-faulted with a
0066  * corresponding layout.
0067  * See also the comment near timens_setup_vdso_data() for details.
0068  */
0069 int vdso_join_timens(struct task_struct *task, struct time_namespace *ns)
0070 {
0071     struct mm_struct *mm = task->mm;
0072     struct vm_area_struct *vma;
0073
0074     mmap_read_lock(mm);
0075     for (vma = mm->mmap; vma; vma = vma->vm_next) {
0076         unsigned long size = vma->vm_end - vma->vm_start;
0077
0078         if (!vma_is_special_mapping(vma, &vvar_mapping))
0079             continue;
0080         zap_page_range(vma, vma->vm_start, size);
0081         break;
0082     }
0083     mmap_read_unlock(mm);
0084     return 0;
0085 }
0086 #else
0087 static inline struct page *find_timens_vvar_page(struct vm_area_struct *vma)
0088 {
0089     return NULL;
0090 }
0091 #endif
0092
0093 static vm_fault_t vvar_fault(const struct vm_special_mapping *sm,
0094                  struct vm_area_struct *vma, struct vm_fault *vmf)
0095 {
0096     struct page *timens_page = find_timens_vvar_page(vma);
0097     unsigned long addr, pfn;
0098     vm_fault_t err;
0099
0100     switch (vmf->pgoff) {
0101     case VVAR_DATA_PAGE_OFFSET:
0102         pfn = virt_to_pfn(vdso_data);
0103         if (timens_page) {
0104             /*
0105              * Fault in VVAR page too, since it will be accessed
0106              * to get clock data anyway.
0107              */
0108             addr = vmf->address + VVAR_TIMENS_PAGE_OFFSET * PAGE_SIZE;
0109             err = vmf_insert_pfn(vma, addr, pfn);
0110             if (unlikely(err & VM_FAULT_ERROR))
0111                 return err;
0112             pfn = page_to_pfn(timens_page);
0113         }
0114         break;
0115 #ifdef CONFIG_TIME_NS
0116     case VVAR_TIMENS_PAGE_OFFSET:
0117         /*
0118          * If a task belongs to a time namespace then a namespace
0119          * specific VVAR is mapped with the VVAR_DATA_PAGE_OFFSET and
0120          * the real VVAR page is mapped with the VVAR_TIMENS_PAGE_OFFSET
0121          * offset.
0122          * See also the comment near timens_setup_vdso_data().
0123          */
0124         if (!timens_page)
0125             return VM_FAULT_SIGBUS;
0126         pfn = virt_to_pfn(vdso_data);
0127         break;
0128 #endif /* CONFIG_TIME_NS */
0129     default:
0130         return VM_FAULT_SIGBUS;
0131     }
0132     return vmf_insert_pfn(vma, vmf->address, pfn);
0133 }
0134
0135 static int vdso_mremap(const struct vm_special_mapping *sm,
0136                struct vm_area_struct *vma)
0137 {
0138     current->mm->context.vdso_base = vma->vm_start;
0139     return 0;
0140 }
0141
0142 static struct vm_special_mapping vvar_mapping = {
0143     .name = "[vvar]",
0144     .fault = vvar_fault,
0145 };
0146
0147 static struct vm_special_mapping vdso64_mapping = {
0148     .name = "[vdso]",
0149     .mremap = vdso_mremap,
0150 };
0151
0152 static struct vm_special_mapping vdso32_mapping = {
0153     .name = "[vdso]",
0154     .mremap = vdso_mremap,
0155 };
0156
0157 int vdso_getcpu_init(void)
0158 {
0159     set_tod_programmable_field(smp_processor_id());
0160     return 0;
0161 }
0162 early_initcall(vdso_getcpu_init); /* Must be called before SMP init */
0163
0164 static int map_vdso(unsigned long addr, unsigned long vdso_mapping_len)
0165 {
0166     unsigned long vvar_start, vdso_text_start, vdso_text_len;
0167     struct vm_special_mapping *vdso_mapping;
0168     struct mm_struct *mm = current->mm;
0169     struct vm_area_struct *vma;
0170     int rc;
0171
0172     BUILD_BUG_ON(VVAR_NR_PAGES != __VVAR_PAGES);
0173     if (mmap_write_lock_killable(mm))
0174         return -EINTR;
0175
0176     if (is_compat_task()) {
0177         vdso_text_len = vdso32_end - vdso32_start;
0178         vdso_mapping = &vdso32_mapping;
0179     } else {
0180         vdso_text_len = vdso64_end - vdso64_start;
0181         vdso_mapping = &vdso64_mapping;
0182     }
0183     vvar_start = get_unmapped_area(NULL, addr, vdso_mapping_len, 0, 0);
0184     rc = vvar_start;
0185     if (IS_ERR_VALUE(vvar_start))
0186         goto out;
0187     vma = _install_special_mapping(mm, vvar_start, VVAR_NR_PAGES*PAGE_SIZE,
0188                        VM_READ|VM_MAYREAD|VM_IO|VM_DONTDUMP|
0189                        VM_PFNMAP,
0190                        &vvar_mapping);
0191     rc = PTR_ERR(vma);
0192     if (IS_ERR(vma))
0193         goto out;
0194     vdso_text_start = vvar_start + VVAR_NR_PAGES * PAGE_SIZE;
0195     /* VM_MAYWRITE for COW so gdb can set breakpoints */
0196     vma = _install_special_mapping(mm, vdso_text_start, vdso_text_len,
0197                        VM_READ|VM_EXEC|
0198                        VM_MAYREAD|VM_MAYWRITE|VM_MAYEXEC,
0199                        vdso_mapping);
0200     if (IS_ERR(vma)) {
0201         do_munmap(mm, vvar_start, PAGE_SIZE, NULL);
0202         rc = PTR_ERR(vma);
0203     } else {
0204         current->mm->context.vdso_base = vdso_text_start;
0205         rc = 0;
0206     }
0207 out:
0208     mmap_write_unlock(mm);
0209     return rc;
0210 }
0211
0212 static unsigned long vdso_addr(unsigned long start, unsigned long len)
0213 {
0214     unsigned long addr, end, offset;
0215
0216     /*
0217      * Round up the start address. It can start out unaligned as a result
0218      * of stack start randomization.
0219      */
0220     start = PAGE_ALIGN(start);
0221
0222     /* Round the lowest possible end address up to a PMD boundary. */
0223     end = (start + len + PMD_SIZE - 1) & PMD_MASK;
0224     if (end >= VDSO_BASE)
0225         end = VDSO_BASE;
0226     end -= len;
0227
0228     if (end > start) {
0229         offset = get_random_int() % (((end - start) >> PAGE_SHIFT) + 1);
0230         addr = start + (offset << PAGE_SHIFT);
0231     } else {
0232         addr = start;
0233     }
0234     return addr;
0235 }
0236
0237 unsigned long vdso_size(void)
0238 {
0239     unsigned long size = VVAR_NR_PAGES * PAGE_SIZE;
0240
0241     if (is_compat_task())
0242         size += vdso32_end - vdso32_start;
0243     else
0244         size += vdso64_end - vdso64_start;
0245     return PAGE_ALIGN(size);
0246 }
0247
0248 int arch_setup_additional_pages(struct linux_binprm *bprm, int uses_interp)
0249 {
0250     unsigned long addr = VDSO_BASE;
0251     unsigned long size = vdso_size();
0252
0253     if (current->flags & PF_RANDOMIZE)
0254         addr = vdso_addr(current->mm->start_stack + PAGE_SIZE, size);
0255     return map_vdso(addr, size);
0256 }
0257
0258 static struct page ** __init vdso_setup_pages(void *start, void *end)
0259 {
0260     int pages = (end - start) >> PAGE_SHIFT;
0261     struct page **pagelist;
0262     int i;
0263
0264     pagelist = kcalloc(pages + 1, sizeof(struct page *), GFP_KERNEL);
0265     if (!pagelist)
0266         panic("%s: Cannot allocate page list for VDSO", __func__);
0267     for (i = 0; i < pages; i++)
0268         pagelist[i] = virt_to_page(start + i * PAGE_SIZE);
0269     return pagelist;
0270 }
0271
0272 static int __init vdso_init(void)
0273 {
0274     vdso64_mapping.pages = vdso_setup_pages(vdso64_start, vdso64_end);
0275     if (IS_ENABLED(CONFIG_COMPAT))
0276         vdso32_mapping.pages = vdso_setup_pages(vdso32_start, vdso32_end);
0277     return 0;
0278 }
0279 arch_initcall(vdso_init);